Project Overview
Vision Based Agent is an autonomous, closed-loop computer-use agent. It interprets screenshots, reasons over them with a multimodal LLM (Google GenAI), and executes precise mouse and keyboard actions via PyAutoGUI to complete tasks — a continuous perceive → reason → act agentic loop.
It's built on LangGraph to orchestrate that cycle. You describe a complex, multi-step task in plain English and it executes across applications, integrating vision models with low-level OS automation (Pillow handles the image side) — a hands-on architecture for turning "see the screen" into "do the work."